On Mon, 19 Sep 2011, Pietro Battiston wrote: By "observation information" gretl means the observation markers in the first column (e.g. dates for time-series, strings for indentifiable units such as countries). No, gretl is not able to guess on that basis. You should be able to do that sort of thing by loading the data series into a matrix. For example, consider the following simplified version of your data: id1 id2 days 1 2 3 1 1 4 2 1 4 3 1 5 1 3 5 2 1 2 3 1 4 The script below will do what you describe, if I've understood you correctly. <hansl> scalar imin = min(id1) scalar imax = max(id1) scalar jmin = min(id2) scalar jmax = max(id2) # count the number of ordered id pairs scalar np = 0 loop i=imin..imax -q loop j=jmin..jmax -q if sum(id1 == i && id2 == j) > 0 np++ endif endloop endloop printf "npairs = %d\n", np matrix C = zeros(np, 3) colnames(C, "id1 id2 days") # cumulate the 'days' data for each ordered id pair scalar r = 0 loop i=imin..imax -q loop j=jmin..jmax -q if sum(id1 == i && id2 == j) > 0 r++ C[r,1] = i C[r,2] = j loop k=1..$nobs -q if id1[k] == i && id2[k] == j C[r,3] += days[k] endif endloop endif endloop endloop print C # write out as gretl data file nulldata np --preserve series id1 = C[,1] series id2 = C[,2] series days = C[,3] store compacted.gdt id1 id2 days </hansl> Allin Cottrell